How We Imagine: Insights from Single Neuron Recordings in the Human Brain

Citation

Wadia, Varun Spenta (2024) How We Imagine: Insights from Single Neuron Recordings in the Human Brain. Dissertation (Ph.D.), California Institute of Technology. doi:10.7907/8nw5-1q14. https://resolver.caltech.edu/CaltechTHESIS:09272023-165737414

Abstract

As human beings interact with the world, we sample it, build representations of its underlying structure, and subsequently use that knowledge for future reasoning - also called modeling the world. This model is generative, allowing our knowledge of the world to influence our perception and in extreme cases induce perception in the absence of any external stimulus. This phenomenon called mental imagery is a remarkable cognitive ability that allows us to remember previous experiences, imagine new ones, make plans, and solve problems. In the visual domain, our ability to generate visual percepts without external stimulation is the basis of our memory for experiences, as episodic memories are simply a subset of all possible experiences we could imagine. Animal studies have yielded rich insight into bottom-up visual processing, from the first discovery of neurons that respond to complex objects like faces to determining the precise code for general objects in macaque inferotemporal (IT) cortex. However, the neural mechanisms of internally generated top-down processing have been much more elusive. Here we present findings on the mechanisms of visual imagery at single neuron resolution. We approached deciphering visual imagery by first laying out coding principles for object perception and then directly comparing responses during viewing to subsequent imagery of those images. We recorded 384 visually responsive neurons in inferotemporal (IT) cortex of 12 epilepsy patients as they viewed and subsequently imagined carefully parametrized visual objects. We verified that neurons in IT cortex are ‘axis tuned’, i.e. as in macaques they represent visual objects by encoding specific axes that span a high dimensional object feature space. 218/384 visually responsive neurons (~58%) were axis tuned, and the axis model explained more variance than other models tested. Armed with this code for visual objects we examined neural responses during pure imagery in the same neurons. We demonstrate robust reactivation of individual neurons across the brain (~35% of neurons across the brain and ~50% of neurons in IT cortex) and a recapitulation of viewing stimulus preference during pure visual imagery in IT. By first uncovering the code for visual objects and examining it during imagery, we demonstrate that neurons in IT cortex subserve visual imagery by reinstating visual context. This study marks the first detailed exploration of visual perception and imagery in the human brain.

Thesis Files